question about "compositional_loss" about deep-image-matting HOT 6 CLOSED

had the same concern!

from deep-image-matting.

Same question!
But the code worked without error.
Do you figure this out? @liminn @mathrho

from deep-image-matting.

@xaibeing Try:

>>>a = np.zeros((2,2,2))
>>>a[:,:,0]      # good
>>>a[:,:,2]      # error
>>>a[:,:,10:20]  # not a error, return : array([], shape=(2, 2, 0), dtype=float64)

from deep-image-matting.

@liminn thanks! you are right, I didn't know array can be used this way.
And a[:,:,10:20] return array[], so I think the compositional_loss didn't take effect, unless fill batch_y with image, fg and bg. right?

from deep-image-matting.

@xaibeing Yes, but maybe the compositional_loss isn't important ^_^

from deep-image-matting.

I found the same issue, interesting bug, LOL!
And it seems that cropped image/alpha/trimap are not resized to 320x320 ......

from deep-image-matting.